Hassibi et al’s (Deepmind) protein folding predictor

I’ve been using the protein folding problem as an example of a really hard problem in computing for a long time: that and real-time weather forecasting have been used by many as part of the case for supercomputers making a real difference.

This new work is an improvement on their 2018 technique (see https://www.nature.com/articles/s41586-019-1923-7.epdf), and is based on deep learning and gradient descent. Now their 2020 technique is an improvement on this (see https://deepmind.com/research/open-source/computational-predictions-of-protein-structures-associated-with-COVID-19 and https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology).

Whys does this matter? Proteins are absolutely central to life on Earth: they are the building blocks of all living entities. Proteins are complex (very complex) molecules made from strings on amino acids, but their behaviour is tightly loud up with their spatial conformation. So if one knows the string of amino acids, one might be able to predict their behaviour. However, their behaviour (what they will react with, how they will change their conformation in electric fields etc.) is very hard to predict from their chemical structure – it needs their conformation as well.

This new advance starts to make determining their structure directly look more possible. And this matters not just for understanding the behaviour of existing proteins, but for predicting the behaviour of synthesised proteins as well.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: